Academic here who has to mark student work.

They learn, approximately, jack shit when using AI incorrectly (I.e. they way the majority of the public use it). AI is a useful tool when used well (e.g. to generate solutions with errors to test knowledge); so far I see little evidence among students it is being well used.